The detection system to predict different types of toxicity. This project uses NLP (TF-IDF) and XGBoost. Set up back-end and front-end servers via Flask, and wrapped them in Docker. Binary classification and multi-targets.
All steps you can see in main file train.ipynb.
- ML: NLP, XGBoost, TF-IDF, sklearn, pandas, numpy, matplotlib
- API: Flask, Flask_wtf
- VM: Docker
- Data from kaggle: Jigsaw Toxic comment competition
Only one feature: comment_text (text)
Feature transform:
- regex clean
- tfidf
1.1 Clone git:
$ git clone https://github.com/hildar/python-flask-docker.git
1.2 If you have some problem with downloading ML model yon can download it from logreg_pipeline.dill or create it from train.ipynb file and put at the folder app/models/;
1.3 Make docker image:
$ cd python-flask-docker/docker
$ docker build -t python-flask-docker app/
1.4 Run docker container:
$ docker run -d -p 8180:8180 -p 8181:8181 python-flask-docker
Now, there are two ways:
Go to the address http://localhost:8180 and use front server. You can manually type some comment at the web form and enjoy the result:
Use Jupiter Notebook "Client.ipynb" and step by step check server located http://localhost:8181.